From Brain-Inspired Intelligence to Evolutionary, Collaborative, and Safe Systems

0 2 minutes read

[Submitted on 31 Mar 2025 (v1), last revised 2 Aug 2025 (this version, v2)]

Authors:Bang Liu, Xinfeng Li, Jiayi Zhang, Jinlin Wang, Tanjin He, Sirui Hong, Hongzhang Liu, Shaokun Zhang, Kaitao Song, Kunlun Zu, Yuheng Cheng, Suyucen Wang, Xiaoqiang Wang , Jiaqi Chen, Huan Zhang, Zhaoyang Yu, Haocen Shi, Boyan Li, Dekun Wu, Fengwei Teng, Xiaojun Jia, Jiawei Xu, Jinyu Xiang, Yizhang Lin, Tianming Liu, Tongliang Liu, Yu SU, Huan Berse, Glain, Qingyun Wu, Yu Gu, Mingchen Zhue, Xinbing Liang, Xiangru Tang, Haohan Wang, Jiaxuan You, Chi Wang, Jian Pei, Qiang Yang, Xiaoliang Qi, ChengLin Wu

View the PDF file from the paper entitled progress and challenges in the foundation agents: From intelligence inspired by the brain to evolutionary, cooperative and safe systems, written by Bang Liu and 47 other books

PDF view

a summary:The appearance of large language models (LLMS) has stimulated a transformational shift in artificial intelligence, which paves the way for advanced smart factors capable of sophisticated thinking, strong perception, and multi -use action across various fields. Since these factors are increasingly pushing artificial intelligence research and practical applications, their design and evaluation and continuous improvement represent complex and multi -faceted challenges. This book provides a comprehensive overview, framing smart factors within the normative brain -inspired structure that merge the principles of cognitive science, neuroscience, and arithmetic research. We are the intention of exploring us into four interconnected parts. First, we systematically achieve the standard basis of smart factors, systematically drawing methodology of their cognitive and operational units on similar human brain functions and clarify the basic components such as memory, modeling of the world, processing rewards, goal, and emotion. Second, we discuss the mechanisms of self -development and adaptation, explore how to improve agents independently their capabilities, adapt to dynamic environments, and achieve continuous learning through automated improvement forms. Third, we study multi -agent systems, and to investigate collective intelligence arising from the agent’s interactions, cooperation, and societal structures. Finally, we address the decisive necessity of building safe and useful artificial intelligence systems, while emphasizing the fundamental and external security threats, moral alignment, durability, and practical mitigation strategies necessary for publication in the real world. By manufacturing the normative AI structure with visions of different disciplines, this poll defines the main challenges and opportunities, which encourages innovations that coordinate technological progress with a meaningful societal benefit.